Pneumatic Muscle Collision Perception And Joint Rotation Control

Robots should not only have high control precision, but also need to possess collision perception ability and flexibility. The robots droved by pneumatic muscles have been widely used for the advantages of light weight, flexibility and good imitation of nature. However, pneumatic muscle has many complex characteristics and it is difficult for people to model and control it or install sensors on its surface. This article mainly studies two problems:One is perceiving radial direction collision based on the differential pressure signals of pneumatic muscle ends; the other is studying the control algorithm of pneumatic muscle joint.This article establishes a platform to test the characteristics of pneumatic muscle and designs a single degree of freedom pneumatic muscle joint. Characteristics of pneumatic muscle and the joint are tested. Pneumatic muscle length-pressure model is established by fitting original curve with two order Gauss function and fitting the hysteresis curve with three orders Fourier function; The test results show that the pneumatic muscle and the joint have strong nonlinearity and hysteresis, which change along with pressure and load.This article puts forward a method that can perceive a radial direction collision (impact) according to the differential pressure signals of pneumatic muscle ends, and distinguishes it from axial impact. Based on the fluid impedance method, the model of axial and radial impact differential pressure signals is established. This article sets up the experimental platform and data acquisition system, studies the impact of load, internal pressure, and impact effect and collision position on the differential pressure signals by experiments. Experimental results show that the amplitude of radial impact differential pressure signals is larger than axial impact under same conditions. Axial phase-frequency curve change promiscuously, while radial phase-frequency curve has periodic variation rule. Four factors all change the amplitude of differential pressure signals, but don’t change the rule of phase-frequency curve. This article uses self-correlation function method to extract the periodic characteristic of phase-frequency curve and distinguishes radial impact from axial impact.This article designs a modified neuron PID control algorithm, defines the gain coefficient of neuron PID control algorithm with the Sigmoid function to improve the self-adaptability and add a attenuation factor in parameter learning process to improve convergence speed. According to the hysteresis reveal from inflation and deflation processes of Pneumatic muscle; this article designed a double structure MFA-DSCMAC control algorithm to learn the charge and discharge processes respectively, which compensates the hysteresis efficiently. A credibility evaluation function is designed to adjust the learning rate, and then damage to neural network caused by sudden interference can be suppressed.This article builds a robotic arm with five degrees of freedom and establishes a control system based on Industrial Personal Computer and C++ software, using the series-wound joint control algorithm to control the robotic arm to complete the pouring water operation.